8635_demo script predictive analysis with hana including pal and r
TRANSCRIPT
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
1/51
DOCUMENT CLASSIFICATION: INTERNAL
General Information: Retail Cross Industry
SAP HANA
Authors: Philip Mugglestone
Date Last Updated: January, 2013
DEMONSTRATION
SCENARIO
PREDICTIVE ANALYTICS
WITH SAP HANA
CUSTOMER SEGMENTATION
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
2/51
2
TABLE OF CONTENTS
TABLE OF CONTENTS ................................................................................................................... 2
1. Demo Script Overview ......................................................................................................... 3
1.1. Demo Description ...................................................................................................................................... 3
1.2.
Protagonists .............................................................................................................................................. 3
1.3.
Business Pain Points................................................................................................................................. 3
1.4.
Key Messages and Value Proposition ...................................................................................................... 4
1.5. Storyflow Summary ................................................................................................................................... 4
2. Technical Requirements ...................................................................................................... 5
2.1.
Prerequisites ............................................................................................................................................. 5
2.2.
System Access Information ....................................................................................................................... 5
2.2.1. System Access ............................................................................................................................. 5
2.2.1.1. SAP Demo Cloud Showroom .................................................................................................... 5
2.2.1.2. Users ......................................................................................................................................... 5
3. Demo Script .......................................................................................................................... 6
3.1. Step-By Step Guide................................................................................................................................... 6
Step 1: Introduction................................................................................................................................ 6
Step 2: Customer Segmentation with SAP HANA PAL ....................................................................... 8
Step 3: Customer Segmentation with R Integration for SAP HANA................................................ 19
Step 4: Recommending the best number of clusters........................................................................ 25
Conclusion ............................................................................................................................................. 30
Step 1: Introduction.............................................................................................................................. 31
Step 2: Single Exponential Smoothing............................................................................................... 33
Step 3: Double Exponential Smoothing.............................................................................................. 39
Step 4: Triple Exponential Smoothing................................................................................................ 45
Step 5: Developer Overview ................................................................................................................. 47
Conclusion 50
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
3/51
3
1.DEMO SCRIPT OVERVIEW
1.1. DEMO DESCRIPTION
The purpose of this demo is to showcase the capabilities of SAP HANA to support predictive analytics in big data
scenarios.
In the first part of this demo, a business user or data analyst (perhaps supported by a data scientist) can perform
customer segmentation in real-time using SAP HANA. The emphasis is on speed and flexibility just seconds are
needed to segment a large dataset and without resorting to sampling the source data. So models are (arguably) more
accurate as all source data is leveraged and the business user wins time back which they can use to perform further
analyses and simulations. In the second part of the demo, a business user or data analyst can perform time series
analysis in real-time using SAP HANA. The emphasis is on speed and flexibility just seconds are needed to
aggregate over a million sales transactions covering 50 electronic items from the last two years and perform focused
time series analysis. Analysts are able to compare actual sales to predicted (i.e fitted) for all products or focused on a
specific product category.
This demo showcases SAP HANA as a platform and is particularly targeted to the SAP HANA developer audience to
demonstrate how Predictive Analytics with SAP HANA can easily be embedded into customer solutions.
This demonstration is based on retail customer data.
However, it can easily be understood by anyone.
The datasets used are hosted by SAP HANA
Predictive analysis is performed by SAP HANA Predictive Analysis Library (PAL) and R Integration for SAP HANA.
Exponential smoothing can be applied to time series data either to produce smoothed data for presentation or in orderto make forecasts.
Single exponential smoothing is the simplest form and easily applied; double exponential smoothing is better suited to
data that trends, whereas triple exponential smoothing also incorporates seasonal changes.
Fit can be fine-tuned via alpha, beta, and gamma factors (,,). Values closer to 0 tend to produce smoother results;
values nearer to 1 give greater weight to recent changes.
NB: A similar scenario is available that combines SAP Predictive Analysis with SAP HANA. SAP Predictive Analysis
provides richer visualizations along with modeling and workflow. The two approaches are complementary. The primary
objective of this demo is to showcase how the predictive capabilities of SAP HANA can be embedded into custom
solutions by developers using SAP HANA as a platform.
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
4/51
4
This demo allows customers and partners to:
Make smarter decisions faster with real-time insights
Accelerate key business processes with rapid analysis and reporting
Invent new business models and processes by leveraging innovative solutions
Reduce TCO with less hardware, maintenance, and testing
There are many use cases for cluster analysis
Customer segmentation
Data reduction / problem refinement when faced with large, complex data sets
Segmenting the market and determining target markets
Product positioning
Selecting test markets
Crime pattern analysis
Medical research, social services, psychiatry, education, archaeology, astronomy, taxonomy
Or the opposite of segmentation anomaly detection
1.4. KEY MESSAGES AND VALUE PROPOSITION
- SAP HANA as a platform to develop custom solutions leveraging all kinds of big data
- Accelerate and improve key business processes through rapid and integrated predictive analysis
- Invent new business models and processes by leveraging innovative predictive analysis solutions
- Exploit SAP HANA to reduce TCO with less hardware, maintenance, and testing
1.5. STORYFLOW SUMMARY
The overall story is:Customer Segmentation
Cluster analysis being used to group or cluster various retail customers given data on recent and total spend,income and age, in order that promotional sales strategies may be applied to groups of similar customers.
Once we have grouped or segmented the customers, we would like to know why customers were allocated toeach segment and when new customers are acquired what group sales strategy they should be assigned to.
We start with an initial data analysis to see if there are any obvious segments / groupings.
Then we run the k-means algorithm to perform the customer segmentation i.e. identify clusters.
When new customers are acquired we score (or predict) which group strategy should be applied to them.
Time Analysis Series
Business analyst wishes to better understand existing sales patterns and predict future sales & revenues, anddo this in an interactive ad-hoc manner for any product category.
Different products exhibit different sales patterns over time. Cameras, for example, are trending
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
5/51
5
2.TECHNICAL REQUIREMENTS
2.1. PREREQUISITES
This demo has been optimized for iPad. For desktops, use of Internet Explorer 9 or Firefox is strongly recommended.
There are no additional pre-requisites.
2.2. SYSTEM ACCESS INFORMATION
2.2.1. S!"#$ A%%#!!
2.2.1.1. SAP D#$& C'&() S*&+,&&$
Showroom =HANA Showroom Edition 2013-1
Click here for information for accessing all SAP Demo Cloud Showrooms
2.2.1.2. U!#,!
Predefined User (Role) Password Component
visual welcome Predictive Analytics
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
6/51
6
Copyright 2012 SAP AG CUSTOMER
3.DEMO SCRIPT
3.1. STEP-BY STEP GUIDE
3.1.1. D#$& /," 1: C(!"&$#, S#0$#"/"&
What to do/say What you should see
Step 1: Introduction
Using Internet Explorer enter:
http://hana1:8020/palr/default.html
logon with VISUAL/welcome
An overview of the scenario is providedon the right hand side of the homescreen.
We have a number of options available well see these in more detail later.
First well get an understanding for thedataset were working with which fornow is a sample of the full dataset withjust 150 retail customers.
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
7/51
7
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on DATA to view the customerdataset
We have data for 150 customers whichincludes information on their:
Lifetime spend (on our productsand services)
New spend (recent spend onour products and services)
Customers annual income
Customer loyalty index (basedon years theyve been acustomer)
Optionally scroll down to show morecustomers.
Also note that the cluster column iscurrently empty.
We are going to segment thesecustomers into a small number of similargroups (clusters) based on thecombination of the 4 key pieces ofinformation outlined above.
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
8/51
8
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Step 2: Customer Segmentation with SAP HANA PAL
Click on HOME to return to the initialscreen
Review the options available:
Source Data: we have just viewed the150 customers dataset (well get to the 1
Million+ shortly)Predictive Library: SAP HANA providestwo options for performing predictiveanalytics and data mining.
PAL is native to SAP HANA andprovides a selection of popularalgorithms for in-memory in-database processing for
optimum performance.
R Integration for SAP HANAprovides the ability to work withover 3500 algorithms availablethrough R
Clusters (K): This is where we selecthow many clusters (or groups orsegments we wish to create)
Recommend K: Well get to this later
NB: Display Data is simply to controlhow many observations (rows) from thesource / result datasets should beshown in the DATA tab. It has no effecton the segmentation process which isperformed on the SAP HANA server.
Press the Perform K-Means Clusteringbutton
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
9/51
9
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
The system will take a second toperform the K-Means processing. As
this is a very small dataset it will be veryfast as youd expect.
The results are displayed in theCLUSTERS tab
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
10/51
10
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
As a first initial analysis of the clusteringyou can see how many customers havebeen assigned to eachcluster/group/segment.
This provides a quick overview of howbalanced the segmentation is.
It's often a good idea to run an algorithmon a small subset of the data to get anidea how the model might work beforeapplying to the larger data set.
Optionally move the mouse over thebars to see more details.
We now want to view the cluster thathas been assigned to each individualcustomer
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
11/51
11
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on DATA to return to thedataset tab
Now we can see that the CLUSTERcolumn has been populated for eachcustomer.
So what? you may be thinking
well weve now seen the process forcustomer segmentation on a tinydataset.
But how will this perform on big data for example similar analysis to segmentmore than 1 million customers?
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
12/51
12
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on HOME to return to the initialscreen
Change the Source Data value to 1Million+ Customers.
Press the Perform K-Means Clusteringbutton
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
13/51
13
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Whilst this runs explain that the K-Means algorithm is a data intensive
process not at all like a simple BIquery.
In fact, K-Means involves multiplepasses of the entiredataset.
It is not generally possible to do really
large data volume data analysis withtraditional systems and the typicalworkaround is to sample (effectivelysubset) the dataset.
However through in-memory computingwith SAP HANA, where the data isstored and processedin SAP HANA
this now becomes possible.
The system needs just a few seconds toperform the K-Means processing!!!
WOW!!!
As previously, the results are displayedin the CLUSTERS tab
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
14/51
14
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Again you can see how many customershave been assigned to each cluster.
Optionally, move the mouse over a barto see more details.
NB: Notice the customer frequencyvalues have increased significantly fromthe previous example now in hundred
on thousands of customers. We reallyare working with over 1 millioncustomers!
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
15/51
15
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on SILHOUETTE to show thesilhouette value
The screen shows the averagesilhouette (or level of confidence in thefit) when 3 clusters are created againstthis dataset.
More explanation of silhouette isprovided in the text on the screen.Basically, the close the value is to 1 themore appropriate the number of clusters(K) will be.
We will come back to this later
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
16/51
16
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on ARCHITECTURE to showthe architecture diagram
So we have seen how complex datamining processing against largevolumes of detailed data becomespossible thanks to SAP HANA.
And its sofastthat you can make thisan iterative and interactive process forexample try a number of different valuesfor K (number of clusters to create) tosee what provides the best fit
At this point, especially to a developeraudience, you can speak to the overallarchitecture of predictive analytics withSAP HANA and how it forms the centerof a rich ecosystem.
Business focused applications such asSAP Predictive Analysis, BI tools, andcustomer applications can all exploit thepower of SAP HANA for predictiveanalysis.
Data can come from any source (notlimited to SAS transactional data).
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
17/51
17
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
To a developer audience, nows a goodtime to explain a little about how easy itis to take advantage of the predictiveanalytics capabilities of SAP HANA
Click on CODE to show the SQLScript
Predictive Analytics processing in SAPHANA is accessed with SAP HANA SQLScript syntax (looks similar to storedprocedures in regular RDBMS).
The procedure is created once duringdevelopment.
In order to execute a K-Means wesimply need to CALL it using the singleline below.
Highlight the pal: lines to show whichPAL algorithms were invoking.
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
18/51
18
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on ARGS to show thearguments that were passed to SAPHANA for processing
A number of parameters are available tocontrol how processing is performed.
These are stored in a HANA table andcan be modified immediately prior toprocessing
In this case the application set theGROUP_NUMBER item to 3 in order tocreate 3 clusters.
The max number of iterations the K-Means algorithm performs can also becontrolled via the MAX_ITERATIONoption etc.
Its really as simple as that with SAPHANA!
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
19/51
19
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Step 3: Customer Segmentation with R Integration for SAP HANA
As well as the Predictive AnalysisLibrary of SAP HANA (PAL) we can alsoleverage R processing from a SAPHANA environment.
More the 3500 algorithms are availablewith R but for this demonstration well
stick with K-Means for simplicity.
Click on HOME to return to the initialscreen
Change the Source Data value to 150Customers.
Change the Predictive Library to RIntegration for SAP HANA.
Change the Clusters (K) value to 10.
Press the Perform K-Means Clusteringbutton
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
20/51
20
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
It runs just like previously however thistime R has been used for the heavylifting to determine the clusters.
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
21/51
21
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on SILHOUETTE to show thesilhouette screen
The screen shows the averagesilhouette (or level of confidence in thefit) when 5 clusters are created againstthis dataset.
Notice that for 5 clusters the silhouettevalue is lower than for 3.
Identifying the best number of clusters(K value) can be a major challengemore to come on this later!
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
22/51
22
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on ARCHITECTURE to show
the architecture diagram
Optionally show the ARCHITECTUREscreen again highlight how RIntegration for SAP HANA fits into theecosystem.
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
23/51
23
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
To a developer audience, nows a good
time to explain a little about how easy itis to take advantage of the predictiveanalytics capabilities of SAP HANA
Click on CODE to show the SQLScript
Predictive Analytics processing in SAPHANA is accessed with SAP HANA SQLScript syntax (looks similar to storedprocedures in regular RDBMS).
The procedure is created once duringdevelopment.
In order to execute a K-Means wesimply need to CALL it using the singleline below.
Highlight the 4 lines between BEGINand END to show the custom R scriptthat was embedded.
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
24/51
24
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on ARGS to show thearguments that were passed to SAPHANA for processing
A number of parameters are available tocontrol how processing is performed.These are stored in a HANA table and
can be modified immediately prior toprocessing
In this case we updated item 1 to 5 inorder to create 5 clusters.
Any R capability can be accessed in this
way from SAP HANA.
Its really as simple as that with SAPHANA!
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
25/51
25
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Step 4: Recommending the best number of clusters
We have already discovered thatidentifying the best number of clusters(K value) can be a major challenge
Wouldnt it be great if the applicationcould find this out for us?
Well the power of in-memory processingin SAP HANA makes it possible!
Lets have SAP HANA performclustering for a range of K values (from2 to 10) in order to determine the bestaverage silhouette value.
Click on HOME to show the homescreen
Change the Predictive Library to SAPHANA Predictive Analysis Library(PAL).
Check the Recommend K box.
Press the Perform K-Means Clusteringbutton
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
26/51
26
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
SAP HANA now performs the k-means
analysis 9 times.
Processing may take a few seconds
and leverages parallel processing todo this extremely quickly!
27
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
27/51
27
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
The SILHOUETTE screen is
displayed
This time the table contains the averagesilhouette value for all 9 K values.
Lets sort the table by silhouette value tosee the largest first.
Click on the SILHOUETTE column
28
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
28/51
28
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Select the Sort Descending option.
29
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
29/51
29
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Now we can see that the first K
displayed is 2 which with the highestaverage silhouette value of 0.6893becomes the recommended K value forthis dataset (based on the 4 customerattributes we used in the K-Meansanalysis).
30
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
30/51
30
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Conclusion
Click on HOME to return to the initialscreen
Lets summarize what we did:
Performed K-Means on 150
customers via SAP HANAPredictive Analysis Library(PAL)
The same for 1 Million+customers took mere seconds!!!
Showed how easy it is fordevelopers to leverage thisfunctionality
Demonstrated R Integration forSAP HANA
Showed how SAP HANAenables new scenarios such asrecommending the best K value
NB: for business analysts
consider SAP PredictiveAnalysis which also works withSAP HANA and providesmodeling and visualizationcapabilities (i.e. no coding)
End of the demo.
31
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
31/51
31
Copyright 2012 SAP AG CUSTOMER
3.1.2 D#$& /," 2: T$# S#,#! A/'!!
What to do/say What you should see
Step 1: Introduction
Before you start, make sure you havethis screen
An overview of the scenario is provided
on the right hand side of the homescreen.
We have a number of options available well see these in more detail later.
First well get an understanding for thedataset were working with whichcontains over a million transactions for a
leading electronic goods store.
32
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
32/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on DATA to view the dataset
We have data for:
Over 1 million transactions
200,000 customers
Around 50 electronic itemsgrouped into 5 productcategories
2 years of history
Optionally scroll down to show moresales transactions.
We are going to perform time seriesanalysis on the purchases by these
customers over the last 2 years - inorder to better understand salespatterns and predict future revenues.
33
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
33/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Step 2: Single Exponential Smoothing
Click on HOME to return to the initial
screen
Review the options available:
Predictive Library: SAP HANA providestwo options for performing predictiveanalytics and data mining.
PAL is native to SAP HANA andprovides a selection of popular
algorithms for in-memory in-database processing foroptimum performance.
R Integration for SAP HANAprovides the ability to work withover 3500 algorithms availablethrough R
In this demo we focus on the PAL.
Product Category: ability to filter basedon product category (Cameras, TVs,etc.)
Exponential smoothing method: type ofsmoothing to perform. Single for now
NB: Display Data is simply to controlhow many observations (rows) from thesource / result datasets should beshown in the DATA tab. It has no effecton the segmentation process which isperformed on the SAP HANA server.
Press the Perform Time SeriesAnalysis button
34
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
34/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
The system will take a second to
perform the time series analysis.
We have not filtered by product categoryso analysis is performed for all of thedata we have over the last 2 years!
This is big data but SAP HANA is able to
process (aggregrate and fit thepredictive model) extremely quickly inreal-time.
And its not necessary to export datainto a separate data mining application the predictive capabilities are native toSAP HANA.
35
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
35/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
The results are displayed in the
PREDICTIONS tab
We see total revenues (in millions) forthe last two years for all products in theblue bars.
The green line shows the fitted
predicted line based on singleexponential smoothing. We will interpretthat shortly
36
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
36/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Optionally you can use the mouse to
highlight information on the chart andsee the actual values.
Sales are rising steadily this is good,but lets now take a look at a specificproduct category.
37
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
37/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on HOME to return to the initial
screen
Change the Product Category value toCameras.
Press the Perform Time SeriesAnalysis button
38
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
38/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
The system needs just a few seconds to
perform the filtering, aggregation, andexponential smoothing analysis!
As previously, the results are displayedin the PREDICTIONS tab
Seems that Cameras follow the overall
trend of steadily rising sales over time.
Notice the green line seems to be a littlelow compared to the chart. This isbecause single exponential smoothinginvolves an inherent time-lag so valuesare typically about one time periodbehind.
NB: It is possible to influence this time-lag with the Start Period slider (not partof the script for this demo).
The analyst is not satisfied with the fit,and would like to consider other
possibilities
39
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
39/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Step 3: Double Exponential Smoothing
Click on HOME to return to the initial
screen
Change the Exponential SmoothingMethod value to Double.
Double exponential smoothing is usuallybetter suited to data that is trending.
Notice that a Trend Smoothing Factorslider has appeared. This allows theuser to fine tune the smoothing process.This is also referred to as the beta ()factor.
Values closer to 0 tend to producesmoother results; values nearer to 1
give greater weight to recent changes.
We will leave the value at 0.1.
Press the Perform Time SeriesAnalysis button
40
Wh t t d / Wh t h ld
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
40/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
This time the fit looks much better!
Also notice that with double exponentialsmoothing we can predict future timeperiods.
Currently were looking 6 months intothe future.
All looks good for Cameras.
But what about TVs?
41
What to do/say What you should see
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
41/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on HOME to return to the initialscreen
Change the Product Category value toTV.
Press the Perform Time SeriesAnalysis button
42
What to do/say What you should see
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
42/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Hmmm the fit is not so good for TVs.
Double exponential smoothing dealswith the trending - but this data appearsto have follow a seasonal cycle.
It seems that sales are lower during thesecond month of each quarter.
Its time to try Triple ExponentialSmoothing
43
What to do/say What you should see
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
43/51
Copyright 2012 SAP AG CUSTOMER
What to do/say What you should see
Click on HOME to return to the initialscreen
Change the Exponential SmoothingMethod value to Triple.
Triple exponential smoothing is usuallybetter suited to data that is both trendingand exhibits seasonality (or any kind ofperiodicity).
Notice that Seasonality radio buttonshave appeared. As our TV sales seemto follow a distinct quarterly cycle wellleave the Quarterly button pressed.
Press the Perform Time Series
Analysis button
44
What to do/say What you should see
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
44/51
Copyright 2012 SAP AG CUSTOMER
at to do say at you s ou d see
Ah thats better. We now have
seasonality included in the model butthe fit of the line is still a little off...
The business analyst suspects thatolder data from last year may be havingundue influence on the fit - perhaps heshould give greater emphasis to morerecent data?
45
What to do/say What you should see
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
45/51
Copyright 2012 SAP AG CUSTOMER
y y
Step 4: Triple Exponential Smoothing
Click on HOME to return to the initialscreen
Notice that a Seasonal SmoothingFactor slider is now present. Thisallows the user to fine tune thesmoothing process. This is also referredto as the gamma () factor.
Values closer to 0 tend to produce
smoother results; values nearer to 1
give greater weight to recent changes.
Change the Seasonal Smoothing
Factor value to 0.4.
At the same time lets project the modelfarther into the future
Change the Forecast Periods value to11.
Press the Perform Time SeriesAnalysis button
46
What to do/say What you should see
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
46/51
Copyright 2012 SAP AG CUSTOMER
Ahh yes!!!
A perfect fit!
Not only do we see the predictionsbetter reflect the recent past, but wealso can now see future revenuesprojected until the end of the summer
next year.
We can continue to work interactively inthis way for any product or category andfor any number of periods into thefuture.
SAP HANA provides not only the in-memory processing power to aggregatethe data in real-time but also to performflexible, interactive, predictive analysisat the same time.
47
What to do/say What you should see
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
47/51
Copyright 2012 SAP AG CUSTOMER
Step 5: Developer Overview
Click on ARCHITECTURE to showthe architecture diagram
So we have seen how complex datamining processing against largevolumes of detailed data becomespossible thanks to SAP HANA.
And its sofastthat you can make thisan iterative and interactive process tosee what provides the best fit
At this point, especially to a developeraudience, you can speak to the overallarchitecture of predictive analytics withSAP HANA and how it forms the centerof a rich ecosystem.
Business focused applications such asSAP Predictive Analysis, BI tools, andcustomer applications can all exploit thepower of SAP HANA for predictiveanalysis.
Data can come from any source (notlimited to SAS transactional data).
48
What to do/say What you should see
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
48/51
Copyright 2012 SAP AG CUSTOMER
To a developer audience, nows a good
time to explain a little about how easy itis to take advantage of the predictiveanalytics capabilities of SAP HANA
Click on CODE to show the SQLScript
Predictive Analytics processing in SAPHANA is accessed with SAP HANA SQLScript syntax (looks similar to storedprocedures in regular RDBMS).
The procedure is created once duringdevelopment.
In order to execute exponentialsmoothing we simply need to CALL itusing the single line below.
Highlight the pal: lines to show whichPAL algorithms were invoking.
49
What to do/say What you should see
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
49/51
Copyright 2012 SAP AG CUSTOMER
Click on ARGS to show the
arguments that were passed to SAPHANA for processing
A number of parameters are available tocontrol how processing is performed.These are stored in a HANA table andcan be modified immediately prior toprocessing
In this case we updated the itemsnamed GAMMA to 0.4 andFORECAST_NUM to 11.
Its really as simple as that with SAPHANA!
50
What to do/say What you should see
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
50/51
Copyright 2012 SAP AG CUSTOMER
Conclusion
Click on HOME to return to the initialscreen
Lets summarize what we did:
Performed filtering, aggregation,and exponential smoothing ontwo years of sales transactionsvia SAP HANA PredictiveAnalysis Library (PAL) inseconds!!!
Single, double, and tripleexponential smoothing ispossible with PAL and suitdifferent data scenarios
encompassing trending andseasonality.
Showed how easy it is fordevelopers to leverage thisfunctionality.
NB: for business analystsconsider SAP PredictiveAnalysis which also works withSAP HANA and providesmodeling and visualizationcapabilities (i.e. no coding)
End of the demo.
51
COPYRIGHT 2412 SAP AG. ALL RIGHTS RESERVED.
-
7/21/2019 8635_Demo Script Predictive Analysis With HANA Including PAL and R
51/51
Copyright 2012 SAP AG CUSTOMER
SAP, R/3, SAP NetWeaver, Duet, PartnerEdge, ByDesign, SAP BusinessObjects Explorer, StreamWork, SAP HANA, and other SAP products and servicesmentioned herein as well as their respective logos are trademarks or registered trademarks of SAP AG in Germany and other countries.
Business Objects and the Business Objects logo, BusinessObjects, Crystal Reports, Crystal Decisions, Web Intelligence, Xcelsius, and other Business Objectsproducts and services mentioned herein as well as their respective logos are trademarks or registered trademarks of Business Objects Software Ltd. BusinessObjects is an SAP company.Sybase and Adaptive Server, iAnywhere, Sybase 365, SQL Anywhere, and other Sybase products and services mentioned herein as well as their respective logosare trademarks or registered trademarks of Sybase Inc. Sybase is an SAP company.
Crossgate, m@gic EDDY, B2B 360, and B2B 360Servi ces are registered trademarks of Crossgate AG in Germany and other countries. Crossgate is an SAPcompany.
All other product and service names mentioned are the trademarks of their respective companies. Data contained in this document serves informational purposesonly. National product specifications may vary.
These materials are subject to change without notice. These materials are provided by SAP AG and its affiliated companies ("SAP Group") for informationalpurposes only, without representation or warranty of any kind, and SAP Group shall not be liable for errors or omissions with respect to the materials. The onlywarranties for SAP Group products and services are those that are set forth in the express warranty statements accompanying such products and services, if any.Nothing herein should be construed as constituting an additional warranty.